DOCSLIB.ORG

Second Order Correctness of Perturbation Bootstrap M-Estimator of Multiple Linear Regression Parameter” (; .Pdf)

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Second Order Correctness of Perturbation Bootstrap M-Estimator of Multiple Linear Regression Parameter” (; .Pdf) Submitted to Bernoulli Second Order Correctness of Perturbation Bootstrap M-Estimator of Multiple Linear Regression Parameter∗ DEBRAJ DASa and S. N. LAHIRIb aDepartment of Statistics, University of Wisconsin-Madison, 1300 University Avenue, Madison, WI 53706, USA. E-mail: [email protected] bDepartment of Statistics, North Carolina State University, 2311 Stinson Dr, Raleigh, NC 27695-8203, USA. E-mail: [email protected] Abstract x′ . Consider the multiple linear regression model yi = iβ + ǫi, where ǫi’s are independent and identically distributed random variables, xi’s are known design vectors and β is the p 1 ׯ vector of parameters. An effective way of approximating the distribution of the M-estimator βn, after proper centering and scaling, is the Perturbation Bootstrap Method. In this current work, second order results of this non-naive bootstrap method have been investigated. Second order correctness is important for reducing the approximation error uniformly to o(n−1/2) to get better inferences. We show that the classical studentized version of the bootstrapped estimator fails to be second order correct. We introduce an innovative modification in the studentized version of the bootstrapped statistic and show that the modified bootstrapped pivot is second order correct (S.O.C.) for approximating the distribution of the studentized M-estimator. Additionally, we show that the Perturbation Bootstrap continues to be S.O.C. when the errors ǫi’s are independent, but may not be identically distributed. These findings establish perturbation Bootstrap approximation as a significant improvement over asymptotic normality in the regression M-estimation. Keywords: M-Estimation, S.O.C., Perturbation Bootstrap, Edgeworth Expansion, Studentization, Residual Bootstrap, Generalized Bootstrap, Wild Bootstrap. 1. Introduction arXiv:1605.01440v2 [math.ST] 17 Dec 2017 Consider the multiple linear regression model : ′ yi = xiβ + ǫi, i =1, 2,...,n (1.1) where y1,...,yn are responses, ǫ1,...,ǫn are independent and identically distributed (IID) random variables with common distribution F (say), x1,..., xn are known non random design vectors and β is the p-dimensional vector of parameters. ∗Research partially supported by NSF grants no. DMS 1310068, DMS 1613192 1 2 Das D. and Lahiri S. N. Suppose β¯ is the M-estimator of β corresponding to the objective function Λ( ) i.e. n · n ′ β¯n = arg min Λ(y x t). Now if ψ( ) is the derivative of Λ( ), then β¯n is the M- t i=1 i − i · · estimator correspondingP to the score function ψ( ) and is defined as the solution of the · vector equation n ′ xiψ(y x β)= 0. i − i Xi=1 It is known [cf. Huber(1981)] that under some conditions on the objective function, design vectors and error distribution F ;(β¯n β) with proper scaling has an asymptotically normal − 2 2 2 2 ′ distribution with mean 0 and dispersion matrix σ Ip where σ = Eψ (ǫ1)/E ψ (ǫ1). After introduction of bootstrap by Efron in 1979 as a resampling technique, it has been widely used as a distributional approximation method. Resampling from the naive empirical distribution of the centered residuals in a regression setup, called residual bootstrap, was introduced by Freedman (1981). Freedman (1981) and Bickel and Freedman (1981b) had ∗ shown that given data, the conditional distribution of √n(βn β¯n) converges to the same − normal distribution as the distribution of √n(β¯n β) when β¯n is the usual least square − estimator of β, that is, when Λ(x)= x2. It implies that the residual bootstrap approximation to the exact distribution of the least square estimator is first order correct as in the case of normal approximation. The advantage of the residual bootstrap approximation over normal approximation for the distribution of linear contrasts of least square estimator for general p was first shown by Navidi (1989) by investigating the underlying Edgeworth Expansion (EE); although heuristics behind the same was given by Liu (1988) in restricted case p = 1. Consequently, EE for the general M-estimator of β was obtained by Lahiri (1989b) when p = 1; whereas the same for the multivariate least square estimator was found by Qumsiyeh (1990a). EE of standardized and studentized versions of the general M-estimator in multiple linear regression setup was first obtained by Lahiri (1992). Lahiri (1992) also established the second order results for residual bootstrap in regression M-estimation. A natural generalization of sampling from the naive empirical distribution is to sample from a weighted empirical distribution to obtain the bootstrap sample residuals. Broadly, the resulting bootstrap procedure is called the weighted or generalized bootstrap. It was introduced by Mason and Newton (1992) for bootstrapping mean of a collection of IID ran- dom variables. Mason and Newton (1992) considered exchangeable weights and established its consistency. Lahiri (1992) established second order correctness of generalized bootstrap in approximating the distribution of the M-estimator for the model (1.1) when the weights are chosen in a particular fashion depending on the design vectors. Wellner and Zhan (1996) proved the consistency of infinite dimensional generalized bootstrapped M-estimators. Con- S.O.C. of Perturbation Bootstrap 3 sequently, Chatterjee and Bose (2005) established distributional consistency of generalized bootstrap in estimating equations and showed that generalized bootstrap can be used in order to estimate the asymptotic variance of the original estimator. Chatterjee and Bose (2005) also mentioned the bias correction essential for achieving second order correctness. An important special case of generalized bootstrap is the bayesian bootstrap of Rubin (1981). Rao and Zhao (1992) showed that the distribution function of M-estimator for the model (1.1) can be approximated consistently by bayesian bootstrap. See the monograph of Barbe and Bertail (2012) for an extensive study of generalized bootstrap. A close relative to the generalized bootstrap procedure is the wild bootstrap. It was in- troduced by Wu (1986) in multiple linear regression model (1.1) with errors ǫi’s being het- eroscedastic. Beran (1986) justified wild bootstrap method by pointing out that the distri- bution of the least square estimator can be approximated consistently by the wild bootstrap approximation. Second order results of wild bootstrap in heteroscedastic regression model was first established by Liu (1988) when p = 1. Liu (1988) also showed that usual residual bootstrap is not capable of approximating the distribution of the least square estimator upto second order in heteroscedastic setup and described a modification in resampling procedure which can establish second order correctness. For general p, the heuristics behind achieving second order correctness by wild bootstrap in homoscedastic least square regression were discussed in Mammen (1993). Recently, Kline and Santos (2011) developed a score based bootstrap method depending on wild bootstrap in M-estimation for the homoscedastic model (1.1) and established consistency of the procedure for Wald and Lagrange Multiplier type tests for a class of M-estimators under misspecification and clustering of data. A novel bootstrap technique, called the perturbation bootstrap was introduced by Jin, Ying, and Wei (2001) as a resampling procedure where the objective function having a U- process structure was perturbed by non-negative random quantities. Jin, Ying, and Wei (2001) showed that in standardized setup, the conditional distribution of the perturbation resampling estimator given the data and the distribution of the original estimator have the same limiting distribution which means this resampling method is first order correct without studentization. In a recent work, Minnier, Tian, and Cai (2011) also applied this perturbation resampling method in penalized regression setup such as Adaptive Lasso, SCAD, lq penalty and showed that the standardized perturbed penalized estimator is first order correct. But, second order properties of this new bootstrap method have remained largely unexplored in the context of multiple linear regression. In this current work, the perturbation bootstrap approximation is shown to be S.O.C. for the distribution of studentized M-estimator for the regression model (1.1). An extension to the case of independent and non-IID errors is also established, showing the robustness of perturbation bootstrap towards the presence 4 Das D. and Lahiri S. N. of heteroscedasticity. Therefore, besides the existing bootstrap methods, the perturbation bootstrap method can also be used in regression M-estimation for making inferences regard- ing the regression parameters and higher order accuracy can be achieved than the normal approximation. A classical way of studentization in bootstrap setup, in case of regression M-estimator and ∗ ∗ ∗−1 ∗ −1 n ′ ∗ for IID errors, is to consider the studentization factor to be σn = snτn , τn = n i=1 ψ (ǫi ), ∗2 −1 n 2 ∗ ∗ ′ ∗ ∗ s = n ψ (ǫ ) where ǫ = y x βn, i 1,...,n , with βn being the perturbationP n i=1 i i i − i ∈{ } bootstrappedP estimator of β, defined in Section 2. Although the residual bootstrapped es- timator is S.O.C. after straight-forward studentization, the same pivot fails to be S.O.C. in the case of perturbation bootstrap. Two important special cases are considered as examples in this respect. The reason behind this failure is that although the bootstrap residuals are sufficient in capturing the variability of the bootstrapped estimator in residual bootstrap, it is not enough in the case of perturbation
Load more

Recommended publications
  • Residuals, Part II
    Biostatistics 650 Mon, November 5 2001 Residuals, Part II Key terms External Studentization Outliers Added Variable Plot — Partial Regression Plot Partial Residual Plot — Component Plus Residual Plot Key ideas/results ¢ 1. An external estimate of ¡ comes from refitting the model without observation . Amazingly, it has an easy formula: £ ¦!¦ ¦ £¥¤§¦©¨ "# 2. Externally Studentized Residuals $ ¦ ¦©¨ ¦!¦)( £¥¤§¦&% ' Ordinary residuals standardized with £*¤§¦ . Also known as R-Student. 3. Residual Taxonomy Names Definition Distribution ¦!¦ ¦+¨-,¥¦ ,0¦ 1 Ordinary ¡ 5 /. &243 ¦768¤§¦©¨ ¦ ¦!¦ 1 ¦!¦ PRESS ¡ &243 9' 5 $ Studentized ¦©¨ ¦ £ ¦!¦ ¤0> % = Internally Studentized : ;<; $ $ Externally Studentized ¦+¨ ¦ £?¤§¦ ¦!¦ ¤0>¤A@ % R-Student = 4. Outliers are unusually large observations, due to an unmodeled shift or an (unmodeled) increase in variance. 5. Outliers are not necessarily bad points; they simply are not consistent with your model. They may posses valuable information about the inadequacies of your model. 1 PRESS Residuals & Studentized Residuals Recall that the PRESS residual has a easy computation form ¦ ¦©¨ PRESS ¦!¦ ( ¦!¦ It’s easy to show that this has variance ¡ , and hence a standardized PRESS residual is 9 ¦ ¦ ¦!¦ ¦ PRESS ¨ ¨ ¨ ¦ : £ ¦!¦ £ ¦!¦ £ ¦ ¦ % % % ' When we standardize a PRESS residual we get the studentized residual! This is very informa- ,¥¦ ¦ tive. We understand the PRESS residual to be the residual at ¦ if we had omitted from the 3 model. However, after adjusting for it’s variance, we get the same thing as a studentized residual. Hence the standardized residual can be interpreted as a standardized PRESS residual. Internal vs External Studentization ,*¦ ¦ The PRESS residuals remove the impact of point ¦ on the fit at . But the studentized 3 ¢ ¦ ¨ ¦ £?% ¦!¦ £ residual : can be corrupted by point by way of ; a large outlier will inflate the residual mean square, and hence £ .
    [Show full text]
  • Testing for Uncorrelated Residuals in Dynamic Count Models with an Application to Corporate Bankruptcy
    Munich Personal RePEc Archive Testing for Uncorrelated Residuals in Dynamic Count Models with an Application to Corporate Bankruptcy Sant’Anna, Pedro H. C. Universidad Carlos III de Madrid May 2013 Online at https://mpra.ub.uni-muenchen.de/48429/ MPRA Paper No. 48429, posted 19 Jul 2013 13:15 UTC Testing for Uncorrelated Residuals in Dynamic Count Models with an Application to Corporate Bankruptcy Pedro H. C. Sant’Anna Universidad Carlos III de Madrid, Getafe, 28903, Spain [email protected] Abstract This article proposes a new diagnostic test for dynamic count models, which is well suited for risk management. Our test proposal is of the Portmanteau-type test for lack of residual autocorrelation. Unlike previous proposals, the resulting test statistic is asymptotically pivotal when innovations are uncorrelated, but not necessarily iid nor a martingale difference. Moreover, the proposed test is able to detect local alter- natives converging to the null at the parametric rate T −1/2, with T the sample size. The finite sample performance of the test statistic is examined by means of a Monte Carlo experiment. Finally, using a dataset on U.S. corporate bankruptcies, we apply our test proposal to check if common risk models are correctly specified. Keywords: Time Series of counts; Residual autocorrelation function; Model checking; Credit risk management. 1 1. Introduction Credit risk affects virtually every financial contract. Therefore the measurement, pricing and management of credit risk have received much attention from economists, bank supervisors and regulators, and financial market practitioners. A widely used measure of credit risk is the probability of corporate default (PD).
    [Show full text]
  • Research Paper Commerce Statistics Inference in Nonlinear Regression Model with Heteroscedastic Errors B. Mahaboob Research Scho
    Volume-5, Issue-5, May - 2016 • ISSN No 2277 - 8160 IF : 3.62 | IC Value 70.36 Commerce Research Paper Statistics Inference in Nonlinear Regression Model With Heteroscedastic Errors B. Mahaboob Research Scholar, Department of Mathematics Dr. M. Ramesh Senior Data Scientist, Tech Mahindra, Hyderabad. Dr. Sk. Khadar Assistant Professor (Senior), Department of Mathematics, VIT, Vellore, Babu Tamilnadu. Academic Consultant, Department of CSE, SPMVV Engineering P. Srivyshnavi College, Tirupati. Prof. P. B.Sireesha Research Scholar, Department of Statistics, S.V.University, Tirupati. Balasiddamuni Rtd. Professor ABSTRACT An advent of modern computer science has made it possible for the applied mathematician to study the inferential aspects of an increasing number of nonlinear regression models in recent years. The inferential aspects for nonlinear regression models and the error assumptions are usually analogous to those made for linear regression models. The tests for the hypotheses on parameters of nonlinear regression models are usually based on nonlinear least squares estimates and the normal assumptions of error variables. A more common problem with data that are best fit by a nonlinear regression model than with data that can be adequately fit by the linear regression model is the problem of heteroscedasticity of errors. The problem of heteroscedasticity can be studied with reference to nonlinear regression models in a variety of ways. In the present study, an attempt has been made by developing inferential method for heteroscedastic nonlinear regression model. KEYWORDS : Intrinsically Nonlinear model; Violation of assumptions on Errors; Problem of Heteroscedasticity. I. INTRODUCTION In estimating the linear regression models, the least squares method of estimation is often applied to estimate the unknown parameters of the linear regression model and errors are usually assumed to be independent and identically distributed normal random variables with zero mean and unknown constant variance.
    [Show full text]
  • A Review on Empirical Likelihood Methods for Regression
    Noname manuscript No. (will be inserted by the editor) A Review on Empirical Likelihood Methods for Regression Song Xi Chen · Ingrid Van Keilegom Received: date / Accepted: date Abstract We provide a review on the empirical likelihood method for regression type inference problems. The regression models considered in this review include parametric, semiparametric and nonparametric models. Both missing data and censored data are accommodated. Keywords Censored data; empirical likelihood; missing data; nonparametric regression; parametric regression; semiparametric regression; Wilks’ theorem. 1 Introduction It has been twenty years since Art Owen published his seminal paper (Owen, 1988) that introduces the notion of empirical likelihood (EL). Since then, there has been a rich body of literature on the novel idea of formulating versions of nonparametric likelihood in various settings of statistical inference. There have been two major reviews on the empirical likelihood. The first review was given by Hall and La Scala (1990) in the early years of the EL method, which summarized some key properties of the method. The second one was the book by the inventor of the methodology (Owen, 2001), which provided a comprehensive overview up to that time. The body of empirical likelihood literature is increasing rapidly, and it would be a daunting task to review the entire field in one review paper like this one. We therefore decided to concentrate our review on regression due to its prominence in statistical S.X. Chen Department of Statistics, Iowa State University, Ames, Iowa 50011-1210, USA and Guanghua School of Management, Peking University, China Tel.: 1-515-2942729 Fax: 1-515-2944040 E-mail: [email protected]; [email protected] I.
    [Show full text]
  • Linear Models with R Julian J.Faraway Downloaded by [University of Toronto] at 16:20 23 May 2014 Texts in Statistical Science
    Downloaded by [University of Toronto] at 16:20 23 May 2014 CHAPMAN & HALL/CRC Texts in Statistical Science Series Series Editors Chris Chatfield, University of Bath, UK Martin Tanner, Northwestern University, USA Jim Zidek, University of British Columbia, Canada Analysis of Failure and Survival Data Peter J.Smith The Analysis and Interpretation of Multivariate Data for Social Scientists David J.Bartholomew, Fiona Steele, Irini Moustaki, and Jane Galbraith The Analysis of Time Series—An Introduction, Sixth Edition Chris Chatfield Applied Bayesian Forecasting and Time Series Analysis A.Pole, M.West and J.Harrison Applied Nonparametric Statistical Methods, Third Edition P.Sprent and N.C.Smeeton Applied Statistics—Handbook of GENSTAT Analysis E.J.Snell and H.Simpson Applied Statistics—Principles and Examples D.R.Cox and E.J.Snell Bayes and Empirical Bayes Methods for Data Analysis, Second Edition Bradley P.Carlin and Thomas A.Louis Downloaded by [University of Toronto] at 16:20 23 May 2014 Bayesian Data Analysis, Second Edition Andrew Gelman, John B.Carlin, Hal S.Stern, and Donald B.Rubin Beyond ANOVA—Basics of Applied Statistics R.G.Miller, Jr. Computer-Aided Multivariate Analysis, Third Edition A.A.Afifi and V.A.Clark A Course in Categorical Data Analysis T.Leonard A Course in Large Sample Theory T.S.Ferguson Data Driven Statistical Methods P.Sprent Decision Analysis—A Bayesian Approach J.Q.Smith Elementary Applications of Probability Theory, Second Edition H.C.Tuckwell Elements of Simulation B.J.T.Morgan Epidemiology—Study Design and Data
    [Show full text]
  • Analysis of the Spread of COVID-19 in the USA with a Spatio-Temporal Multivariate Time Series Model
    International Journal of Environmental Research and Public Health Article Analysis of the Spread of COVID-19 in the USA with a Spatio-Temporal Multivariate Time Series Model Rongxiang Rui 1 , Maozai Tian 2 , Man-Lai Tang 3,* , George To-Sum Ho 4 and Chun-Ho Wu 4 1 School of Statistics, Renmin University of China, Beijing 100872, China; [email protected] 2 College of Medical Engineering and Technology, Xinjiang Medical University, Ürümqi 830011, China; [email protected] 3 Department of Mathematics, Statistics and Insurance, Hang Seng University of Hong Kong, Hong Kong, China 4 Department of Supply Chain and Information Management, Hang Seng University of Hong Kong, Hong Kong, China; [email protected] (G.T.-S.H.); [email protected] (C.-H.W.) * Correspondence: [email protected] Abstract: With the rapid spread of the pandemic due to the coronavirus disease 2019 (COVID-19), the virus has already led to considerable mortality and morbidity worldwide, as well as having a severe impact on economic development. In this article, we analyze the state-level correlation between COVID-19 risk and weather/climate factors in the USA. For this purpose, we consider a spatio-temporal multivariate time series model under a hierarchical framework, which is especially suitable for envisioning the virus transmission tendency across a geographic area over time. Briefly, our model decomposes the COVID-19 risk into: (i) an autoregressive component that describes the within-state COVID-19 risk effect; (ii) a spatiotemporal component that describes the across-state COVID-19 risk effect; (iii) an exogenous component that includes other factors (e.g., weather/climate) that could envision future epidemic development risk; and (iv) an endemic component that captures the function of time and other predictors mainly for individual states.
    [Show full text]
  • Power Maximization and Size Control in Heteroskedasticity and Autocorrelation Robust Tests with Exponentiated Kernels
    POWER MAXIMIZATION AND SIZE CONTROL IN HETEROSKEDASTICITY AND AUTOCORRELATION ROBUST TESTS WITH EXPONENTIATED KERNELS By Yixiao Sun, Peter C. B. Phillips and Sainan Jin January 2010 COWLES FOUNDATION DISCUSSION PAPER NO. 1749 COWLES FOUNDATION FOR RESEARCH IN ECONOMICS YALE UNIVERSITY Box 208281 New Haven, Connecticut 06520-8281 Power Maximization and Size Control in Heteroskedasticity and Autocorrelation Robust Tests with Exponentiated Kernels Yixiao Sun Department of Economics University of California, San Diego Peter C. B. Phillips Yale University, University of Auckland, Singapore Management University & University of Southampton Sainan Jin School of Economics Singapore Management University, August 14, 2009 The authors gratefully acknowledge partial research support from NSF under Grant No. SES- 0752443 (Sun), SES 06-47086 (Phillips), and NSFC under grant No. 70501001 and 70601001 (Jin). ABSTRACT Using the power kernels of Phillips, Sun and Jin (2006, 2007), we examine the large sample asymptotic properties of the t-test for di¤erent choices of power parameter (). We show that the nonstandard …xed- limit distributions of the t-statistic provide more accurate approximations to the …nite sample distributions than the conventional large- limit distribution. We prove that the second-order corrected critical value based on an asymptotic expansion of the nonstandard limit distribution is also second- order correct under the large- asymptotics. As a further contribution, we propose a new practical procedure for selecting the test-optimal power parameter that addresses the central concern of hypothesis testing: the selected power parameter is test-optimal in the sense that it minimizes the type II error while controlling for the type I error. A plug-in procedure for implementing the test-optimal power parameter is suggested.
    [Show full text]
  • Optimal Bandwidth Selection in Heteroskedasticity#Autocorrelation
    Optimal Bandwidth Selection in Heteroskedasticity-Autocorrelation Robust Testing Yixiao Sun Department of Economics University of California, San Diego Peter C. B. Phillips Cowles Foundation, Yale University, University of Auckland & University of York Sainan Jin Guanghua School of Management Peking University January 5, 2011 Abstract The paper considers studentized tests in time series regressions with nonparamet- rically autocorrelated errors. The studentization is based on robust standard errors with truncation lag M = bT for some constant b (0; 1] and sample size T: It is shown that the nonstandard …xed-b limit distributions of2 such nonparametrically studentized tests provide more accurate approximations to the …nite sample distributions than the standard small-b limit distribution. We further show that, for typical economic time series, the optimal bandwidth that minimizes a weighted average of type I and type II errors is larger by an order of magnitude than the bandwidth that minimizes the asymptotic mean squared error of the corresponding long-run variance estimator. A plug-in procedure for implementing this optimal bandwidth is suggested and simula- tions (not reported here) con…rm that the new plug-in procedure works well in …nite samples. Keywords: Asymptotic expansion, bandwidth choice, kernel method, long-run variance, loss function, nonstandard asymptotics, robust standard error, Type I and Type II errors An earlier version of the paper is available as Sun, Phillips and Jin (2006). The present version is a substantially shorter paper and readers are referred to the earlier version for further details, discus- sion, proofs, and some asymptotic expansion results of independent interest. The authors thank Whitney Newey, two anonymous referees, and many seminar participants for comments on earlier versions.
    [Show full text]
  • Risk Segmentation Using Bayesian Quantile Regression with Natural Cubic Splines
    Open Access Austin Statistics A Austin Full Text Article Publishing Group Review Article Risk Segmentation Using Bayesian Quantile Regression with Natural Cubic Splines Xia M* Abstract Department of Statistics, Northern Illinois University, USA An insurance claims department is often interested in obtaining the possible *Corresponding author: Xia M, Division of Statistics, distribution of a claim for the purpose of risk management. Information such as Northern Illinois University, 300 Normal Road, DeKalb, the probability that a claim will exceed a certain amount is helpful for matching IL 60115, USA, Tel: 815-753-6795; Fax: 815-753-6776; the claim complexity with the specialty of claim adjusters. Using information Email: [email protected] available on the claim and the parties involved, we propose a Bayesian quantile regression model for the purpose of risk identification and segmentation in the Received: May 29, 2014; Accepted: July 11, 2014; claims department. Natural cubic splines are used in order to estimate a smooth Published: July 16, 2014 relationship between the expected quantiles and continuous explanatory variables such as the age of the claimant. A case study is conducted using the Medical Large Claims Experience Study data from the Society of Actuaries. For the claimant age factor that we study, we observe that the high-risk groups, such as the infants and elderly, exhibit a much higher risk in terms of high quantiles (such as the 99% and 99.5% percentiles) than that is revealed by the mean or the median. Particularly for the claims data where there are various characteristics available on the claimant and other parties involved, our model may reveal helpful information on the possible extremal risk that may be under looked in traditional claims modeling.
    [Show full text]
  • Gelman and Shalizi, 2013
    8 British Journal of Mathematical and Statistical Psychology (2013), 66, 8–38 © 2012 The British Psychological Society www.wileyonlinelibrary.com Philosophy and the practice of Bayesian statistics ∗ Andrew Gelman1 and Cosma Rohilla Shalizi2 1Department of Statistics and Department of Political Science, Columbia University, New York, USA 2Statistics Department, Carnegie Mellon University, Santa Fe Institute, Pittsburgh, USA A substantial school in the philosophy of science identifies Bayesian inference with inductive inference and even rationality as such, and seems to be strengthened by the rise and practical success of Bayesian statistics. We argue that the most successful forms of Bayesian statistics do not actually support that particular philosophy but rather accord much better with sophisticated forms of hypothetico-deductivism. We examine the actual role played by prior distributions in Bayesian models, and the crucial aspects of model checking and model revision, which fall outside the scope of Bayesian confirmation theory. We draw on the literature on the consistency of Bayesian updating and also on our experience of applied work in social science. Clarity about these matters should benefit not just philosophy of science, but also statistical practice. At best, the inductivist view has encouraged researchers to fit and compare models without checking them; at worst, theorists have actively discouraged practitioners from performing model checking because it does not fit into their framework. 1. The usual story – which we don’t like In so far as I have a coherent philosophy of statistics, I hope it is ‘robust’ enough to cope in principle with the whole of statistics, and sufficiently undogmatic not to imply that all those who may think rather differently from me are necessarily stupid.
    [Show full text]
  • 7 Detection and Modeling of Nonconstant Variance
    CHAPTER 7 ST 762, M. DAVIDIAN 7 Detection and modeling of nonconstant variance 7.1 Introduction So far, we have focused on approaches to inference in mean-variance models of the form 2 2 E(Yj|xj)= f(xj, β), var(Yj|xj)= σ g (β, θ, xj) (7.1) under the assumption that we have already specified such a model. • Often, a model for the mean may be suggested by the nature of the response (e.g., binary or count), by subject-matter theoretical considerations (e.g., pharmacokinetics), or by the empirical evidence (e.g., models for assay response). • A model for variance may or may not be suggested by these features. When the response is binary, the form of the variance is indeed dictated by the Bernoulli distribution, while for data in the form of counts or proportions, for which the Poisson or binomial distributions may be appropriate, the form of the variance is again suggested. One may wish to consider the possibility of over- or underdispersion in these situations; this may reasonably be carried out by fitting a model that accommodates these features and determining if an improvement in fit is apparent using methods for inference on variance parameters we will discuss in Chapter 12. Alternatively, when the response is continuous (or approximately continuous), it is often the situation that there is not necessarily an obvious relevant distributional model. As we have discussed in some of the examples we have considered, several sources of variation may combine to produce patterns that are not well described by the kinds of variance models dictated by popular distributional assumptions such as the gamma or lognormal distributions.
    [Show full text]
  • Autocorrelation-Robust Inference*
    G. S. Maddala and C. R. Rao, eds., ok of Statistics, Vol. 15 © 1997 EIsevier Science B.V. AH rights reserved 11 Autocorrelation-Robust Inference* P. M. Robinson and C. Ve/asco 1. Introduction Time series data occur commonly in the natural and engineering sciences, eco­ nomics and many other fields of enquiry. A typical feature of such data is their apparent dependence across time, for example sometimes records close together in time are strongly correlated. Accounting for serial dependence can con­ siderably complicate statistical inference. The attempt to model the dependence parametrically, or even nonparametrically, can be difficult and computationally expensive. In sorne circumstances, the serial dependence is merely a nuisance feature, interest focusing on "static" aspects such as a location parameter, or a probability density function. Here, we can frequently base inference on point or function estimates that are natural ones to use in case of independence, and may well be optimal in that case. Such estimates will often be less efficient, at least asymp­ totically, than ones based on a comprehensive model that incorporates the serial dependence. But apart from their relative computational simplicity, they often remain consistent even in the presence of certain forms of dependence, and can be more reliable than the "efficient" ones, which can sometimes become inconsistent when the dependence is inappropriately dealt with, leading to statistical inferences that are invalid, even asymptotically. In this sense the estimates can be more "robust" than "efficient" ones, and the main focus in this paper is their use in statistical inference, though we also discuss attempts at more efficient inference.
    [Show full text]